基于动态共现的中文话题关联检测  

CHINESE STORY LINK DETECTION BASED ON DYNAMIC CO-OCCURRENCE

作　　者：庞海杰[1] 

机构地区：[1]青岛滨海学院大专理科基础学院,山东青岛266555

出　　处：《计算机应用与软件》2012年第3期115-117,共3页Computer Applications and Software

基　　金：国家自然科学基金项目(60773034)

摘　　要：话题关联检测是话题检测与跟踪的一项子任务,是判断随机抽取的两篇新闻报道是否讨论同一个话题的技术。受词语共现模型的启发,结合话题关联检测的特点,提出了词语间的动态同现关系,实现了基于动态共现关系的报道相似度计算方法;探讨了相似度计算方法在中文话题关联检测中的应用。通过实验可知,动态共现关系可以在一定程度上反映报道的语义信息,相似度计算方法很好地改善了中文话题关联检测系统的性能,取得了不错的效果。Story link detection is a subtask of topic detection and tracking.It is a technology to judge whether two randomly selected news stories are discussing a same event.Motivated by the word co-occurrence model,by integrating characteristics of story link detection,the paper proposes a dynamic co-occurrence relationship among words and realizes a story similarity computation method based on dynamic co-occurrence.Then the application of the similarity computation method to Chinese story link detection is discussed.Experimental results show that dynamic co-occurrence can express the semantic information of a story to a certain degree.The similarity computation method improves a lot the performance of the Chinese story link detection system.There have been good feedbacks.

关 键 词：话题关联检测 话题检测与跟踪 动态共现 归一化检测开销 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]